Analysis of the Isomap projection of the Cytokine compendium dataset. On the Panel A the cytokine compendium dataset was projected into the low-dimensional subspace using Isomap. Next, clusters were identified using the Expectation Maximisation (EM) clustering algorithm with the number of clusters set to 6 (clusters are labelled with different symbols). Panels B-C: Graph-based clustering of the two-dimensional Isomap embedding using the same parameters as for graph-based clustering in Figure 4, but with the expected number of clusters increased from 7 to 9, original clusters remain intact and partitioning is made across the super cluster. When the number of clusters is 9 the Insulin/Insulin-TNF high concentration cluster is partitioned in separate Insulin and Insulin-TNF clusters.